Regime-Augmented HARQ

• Regime-Augmented HARQ is an adaptive method that modifies power, rate, and retransmission strategies based on multi-bit feedback and channel statistics.
• It employs dynamic and geometric programming techniques to optimize resource allocation, minimizing outage probability under power and latency constraints.
• The approach demonstrates significant SNR gains in fading channels and is pivotal for applications requiring ultra-reliable, low-latency wireless communications.

Regime-augmented Hybrid Automatic Repeat Request (HARQ) designates any HARQ protocol in which the operation—including power, rate, retransmission structure, and queueing—adapts to a dynamically estimated "regime" based on channel statistics, instantaneous or accumulated feedback, or system-level constraints such as reliability, latency, or resource efficiency. Pioneered in frameworks that leverage multi-bit, quantized, or otherwise enriched feedback beyond the conventional one-bit ACK/NACK, regime-augmented HARQ generalizes canonical HARQ schemes to a family whose protocols, resource allocation, and feedback granularity are adaptively tailored for optimal performance across operating regimes (Jabi et al., 2013, Trillingsgaard et al., 2018, Sassioui et al., 2016, Dosti et al., 2020, Nadeem et al., 2021, Dinh et al., 2022).

1. System Models and Regime Representation

The regime-augmented HARQ paradigm assumes a block-fading channel model, typically single-input single-output, with up to $K$ transmission attempts per packet (truncated HARQ). The $k$th transmission round observes

$$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$

where $\mathbf x_k$ is a unit-power codeword, $\gamma_k$ the instantaneous SNR (independent across blocks, statistics known at transmitter), and $\mathbf z_k$ is AWGN. Critically, $P_k(\mathsf{fb}_{k-1})$—the transmit power or, more generally, resource allocation in the $k$th round—is chosen as a function of the multi-bit feedback $\mathsf{fb}_{k-1}$ enumerating the regime.

Feedback alphabets vary:

• Standard HARQ: $$\mathsf{fb}_{k-1}\in\{0\equiv\text{NACK},\,1\equiv\text{ACK}\}$$.
• Regime-augmented: Receiver quantizes a scalar decoder state (e.g., accumulated mutual information $k$0, instantaneous SNR $k$1, or backlog of unresolved bits $k$2) into $k$3 regimes: $k$4, potentially encoding queue state, channel statistics, or decoding reliability estimate as well (Jabi et al., 2013, Trillingsgaard et al., 2018).

Decoding is governed by either:

• Incremental redundancy (IR): $k$5.
• Chase combining (CC): $k$6.

Outage is declared if $k$7 (IR) or $k$8 (CC).

Regime thus encapsulates a sufficient statistic for optimizing future transmissions: past ACK/NACKs, quantized decoder state, or broader traffic/channel context.

2. Optimization Frameworks for Regime Adaptation

Regime-augmented HARQ optimization targets minimization of final outage probability $k$9 under resource constraints:

$$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$0

subject to

$$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$1

The admissible set of power or rate policies can be formulated as a Markov (or semi-Markov) decision process since the regime-encoding feedback $$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$2 can summarize all past observed CSI and/or decoder history (Jabi et al., 2013, Trillingsgaard et al., 2018). The typical solution methods are:

Dynamic Programming (General SNR): The cost-to-go function $$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$3 is backward recursed over transmitter state $$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$4 (quantized feedback), accounting at every stage for both expected reliability cost and resource penalty, with $$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$5 as the Lagrange multiplier for enforcing power constraints.

Geometric Programming (High-SNR Regime): At high SNR, outage and average power are monomials in $$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$6. For Nakagami-$$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$7 fading: $$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$8 can be cast as a standard geometric program, whose solution yields explicit allocation laws: $$\mathbf y_k = \sqrt{\gamma_k\,P_k(\mathsf{fb}_{k-1})}\;\mathbf x_k + \mathbf z_k$$9 reflecting the power-law relationship between regime/round index and allocated power (Jabi et al., 2013).

3. Feedback Granularity, Power/Rate Laws, and Protocol Variants

Increasing feedback resolution (i.e., more finely quantized regimes) monotonically improves protocol performance, but returns diminish rapidly beyond 4–5 regime levels (Jabi et al., 2013, Trillingsgaard et al., 2018). Regime signals may encode:

• $\mathbf x_k$0 quantized mutual information or decoder state $\mathbf x_k$1 (multi-bit regime-augmented HARQ) (Jabi et al., 2013);
• $\mathbf x_k$2 quantized channel state or backlog for Expandable Message Space (EMS) protocols (Trillingsgaard et al., 2018);
• $\mathbf x_k$3 instantaneous SNR or error states for regime-adaptive switching (e.g., between HARQ and AMC, or selection of sub-codeword lengths) (Sassioui et al., 2016).

Other protocol variants include:

• EMS/BRQ: The EMS model appends new bits continuously, jointly decoded when backlog permits. With infinite feedback, the optimal rate is aligned to $\mathbf x_k$4 at each slot, matching the BRQ (Backtrack Retransmission Request) throughput. Limited-feedback EMS—by quantizing the regime—retains nearly all the throughput advantage of BRQ with just three or four regime levels (Trillingsgaard et al., 2018).
• Packet-dropping and Variable-length HARQ: In fast fading, regime adaptation by dropping HARQ packets and restarting AMC (if the SNR regime improves) or by dynamically optimizing codeword lengths (VL-HARQ) recaptures much of the AMC’s advantage while limiting HARQ's rate-locking penalty (Sassioui et al., 2016).

4. Performance Gains and Scaling Laws

Rigorous analysis and numerical results demonstrate substantive performance improvement of regime-augmented HARQ over conventional constant-power or single-bit-feedback HARQ, e.g.:

• With Nakagami-2 fading, $\mathbf x_k$5, $\mathbf x_k$6, the regime-augmented protocol yields $\mathbf x_k$75 dB SNR gain at outage $\mathbf x_k$8 compared to constant power, with about $\mathbf x_k$9 dB attributable to power allocation and a further $\gamma_k$0 dB from regime-based adaptation (Jabi et al., 2013).
• Open-loop, one-shot transmission to ultra-reliable outage targets (e.g. $\gamma_k$1) is power-prohibitive ($\gamma_k$2 dB SNR for typical rates and blocklengths), while regime-augmented HARQ cuts average and peak power by $\gamma_k$3–$\gamma_k$4 dB for IR-HARQ with $\gamma_k$5 rounds (Dosti et al., 2020).
• IR regime-augmented HARQ consistently yields the largest power reductions, followed by CC and then Type-I ARQ. Power allocation laws are strictly increasing in round index (late rounds get more power), and are optimal under convexity/KKT analysis (Dosti et al., 2020).
• In URLLC and low-latency settings, non-orthogonal HARQ (N-HARQ) and proactive regime-augmented protocols, leveraging parallel redundancy or time/power-sharing, achieve $\gamma_k$6 lower delivery delay and $\gamma_k$7 dB power savings over conventional HARQ, while closely matching reliability or throughput constraints (Nadeem et al., 2021, Dinh et al., 2022).

Protocol	Feedback Regimes	Outage Diversity	SNR Gain vs. Constant	Complexity
Constant-power HARQ	1-bit	$\gamma_k$8	baseline	low
Regime-aug. (multi-bit)	$\gamma_k$9-bit	$\mathbf z_k$0	up to 5 dB	moderate
EMS/BRQ	$\mathbf z_k$1-bit	optimal	near capacity	high (joint dec)
N-HARQ	$\mathbf z_k$2-bit	equal to O-HARQ	1 dB at~0.7 throughput	high (SIC, FBL)

5. Practical Design Guidelines and Regime-Adaptive Policy Synthesis

Regime-augmented HARQ system design proceeds by:

1. Regime Quantization: Map decoder or channel statistics to a regime index with $\mathbf z_k$3–$\mathbf z_k$4 bins to approach optimal trade-offs between complexity, control overhead, and performance (Jabi et al., 2013, Trillingsgaard et al., 2018).
2. Policy Synthesis: Formulate the regime-indexed resource allocation policy as a dynamic program or, at high SNR, via explicit geometric program or KKT solution (Jabi et al., 2013, Dosti et al., 2020). Use EMS/BRQ rules if delayed CSIT is available and strict zero-outage is targeted (Trillingsgaard et al., 2018).
3. Cross-layer Adaptation: For rapidly-varying channels (fast fading), apply regime-switching policies such as packet-dropping HARQ (PD-HARQ) or variable-length HARQ (VL-HARQ), or revert to AMC with outer ARQ, preferentially based on regime estimate (Sassioui et al., 2016).
4. Parallel/Proactive Mechanisms: In latencysensitive, URLLC, or high-traffic regimes, deploy proactive or non-orthogonal HARQ with cluster-based parallel redundancy controlled by regime-adapted Lyapunov optimization, tuning parameters for required reliability and resource efficiency (Dinh et al., 2022, Nadeem et al., 2021).
5. Power Allocation Law: Always employ strictly increasing power profiles; allocate minimal power in early rounds, ramping up as necessary, with target outage, maximum rounds, and rate determining the law parameters (Dosti et al., 2020).

6. Perspectives and Extensions

The regime-augmented HARQ approach forms a unifying principle for cross-layer wireless reliability under power, latency, and queueing constraints. Fundamental insights include:

• Finer regime quantization gives diminishing returns after $\mathbf z_k$5.
• In the finite blocklength, ultra-reliable domain (URLLC), regime-augmented HARQ is essential for meeting power and latency targets.
• For time-varying and resource-constrained radio access networks, regime-augmented policies controlled by Lyapunov or Markov-queue frameworks yield near-optimal tradeoffs between latency, reliability, and resource efficiency, as demonstrated in proactive HARQ implementations (Dinh et al., 2022).

A plausible implication is that as wireless standards evolve, regime-augmented HARQ will increasingly serve as the design foundation, with protocol variants tailored to operating points of feedback overhead, hardware complexity, and stringent reliability/latency guarantees. Further integration with scheduling, buffer management, and cross-layer queueing policies remains an active research direction (Jabi et al., 2013, Dinh et al., 2022).